Mistral doesn’t have a speech model, until now. Voxtral TTS, released April 3, 2026, is the company’s first text-to-speech model and it’s open-source. That means teams can download it, run it locally, and build products on top of it without a licensing fee or a cloud API call.
Nine languages are confirmed supported: English, French, German, Spanish, Dutch, Portuguese, Italian, Hindi, and one additional language not specified in available source content. According to TechCrunch’s reporting, the model is open-source and multilingual, with Mistral’s official announcement describing it as delivering “state-of-the-art performance in multilingual voice generation.” That characterization is Mistral’s own, no independent benchmarks have evaluated the model yet.
Mistral states Voxtral TTS outperforms Whisper large-v3 on speech generation tasks. Whisper large-v3, also an open-source model, is widely used for speech transcription and is a meaningful comparison point in the open-source audio space. That comparison is self-reported, Mistral’s claim, not a third-party finding. Low-latency real-time voice output is also part of Mistral’s stated design intent, though this hasn’t been independently evaluated. As an open-source model, Voxtral TTS can be deployed locally, reducing reliance on cloud infrastructure, though specific optimization for edge hardware hasn’t been confirmed in available source material.
The significance isn’t hard to find. Voice AI has largely been dominated by proprietary APIs, ElevenLabs, OpenAI’s TTS, and similar services that require sending audio generation requests to external servers. A capable, open-source, multilingual TTS model changes the calculus for developers building anything with a privacy requirement, a latency constraint, or a need for offline operation. It also changes the calculus for teams building in markets where English isn’t the primary language.
Mistral has built its brand on open-weight releases, and Voxtral TTS fits that pattern. The company has consistently released models that can run without cloud dependency. Each release expands what’s possible without a platform subscription. The speech gap in the open-source AI stack has been real, text and vision models have been well-served by open-weight options, but high-quality TTS lagged. Voxtral TTS is the first serious open-source entry from a major AI lab in that gap.
This release pairs naturally with the Gemma 4 announcement from the same day. Google’s multimodal model handles text and vision locally. Mistral’s model handles speech locally. A local AI stack covering text, vision, and voice is more complete today than it was last week.
Watch for independent quality evaluations. The self-reported comparison to Whisper large-v3 will be tested quickly, the open-source community benchmarks these models fast. Also watch for the specific open-source license details, which weren’t confirmed in available source material. The license terms matter for commercial use cases. Once those are clear, enterprise adoption decisions can proceed with full information.